Abstract: Lithology identification while drilling is an important geological guarantee means for transparent detection of coal mine geology. The traditional lithology identification method mainly relies on manual judgment, which relies on the accumulation of experience and professional knowledge and is subjectively affected. In recent years, intelligent lithology identification methods have emerged, which use machine learning lithology recognition models to intelligently identify lithology, and the accuracy of lithology recognition by machine learning is higher than that of manual single drilling parameters, but there is room for improvement. Based on this, this paper upgrades the comprehensive measurement system while drilling on the basis of the crawler full hydraulic tunnel drilling rig. Rock formations with different lithologic combinations were tested while drilling. A two-parameter lithology discrimination system combining drilling parameters and natural gamma was established. In view of the shortcomings of traditional algorithms such as support vector machine, such as linear weight matrix, large number of required parameters, and limited feature extraction ability, the KAN network was applied to lithology intelligent identification. The results show that for the four machine learning algorithms, such as KNN, SVM, DT and KAN, the two-parameter discrimination system using drilling parameters and natural gamma can significantly improve the accuracy of lithology identification compared with the single-parameter discrimination method of drilling parameters or gamma parameters. In terms of machine learning algorithms, the KAN network improves the accuracy compared with the other three traditional machine learning methods, which provides an effective method for accurately identifying the lithology of coal-bearing strata.